Downloaded By: [LCMR - Journal of Cardiovascular Magnetic Resonance] At: 23:31 29 March 2 Journal of Cardiovascular Magnetic Resonance (2007) 9, 525–537 Copyright c 2007 Informa Healthcare ISSN: 1097-6647 print / 1532-429X online DOI: 10.1080/10976640601187604 Imaging Sequences for First Pass Perfusion—A Review Peter Kellman, PhD and Andrew E. Arai, MD Laboratory of Cardiac Energetics, National Heart, Lung and Blood Institute, National Institutes of Health, Department of Health and Human Services, Bethesda, Maryland, USA ABSTRACT Myocardial perfusion imaging sequences and analysis techniques continue to improve. We review the state-of-the-art in cardiovascular magnetic resonance first pass perfusion pulse se- quences including the application of parallel imaging. There are a wide range of sequence designs and parameters to consider when optimizing an acquisition protocol. The interdepen- dence of these parameters forces the user to make compromises. We describe the technical issues and provide insights into the various performance tradeoffs. We also review the basic de- sign for T1-weighted first pass myocardial perfusion imaging and go on to discuss the tradeoffs associated with various schemes to provide multi-slice coverage. Artifact mechanisms are dis- cussed and related to sequence design and parameters. The selection of quantitative versus qualitative analysis affects various performance requirements, such as spatial and temporal resolution and linearity of enhancement. Understanding the interaction between the pulse se- quence parameters and resulting image quality is important for improving myocardial perfusion imaging. INTRODUCTION Myocardial perfusion imaging using cardiovascular magnetic resonance (CMR) has undergone steady improvement since it was first proposed over 20 years ago (1, 2). Yet there remain challenges for widespread clinical acceptance, and there is still considerable debate as to which technique is best. We review the state-of-the-art first pass perfusion pulse sequences including the application of parallel imaging for accelerated acquisition. Per- formance optimization is difficult due to the large number of parameters affecting image acquisition. Furthermore, the inter- dependence of these parameters forces the user to make com- promises. As the field has not come to clear methodological Keywords: Myocardial Perfusion, Cardiovascular MR, Ischemia, Parallel Imaging. Supported by a grant from the National Heart, Lung, and Blood Institute, National Institutes of Health, Intramural Research Program Correspondence to: Peter Kellman Laboratory of Cardiac Energetics National Institutes of Health National Heart, Lung and Blood Institute 10 Center Drive, MSC-1061 Building 10, Room B1D416 Bethesda, MD 20892-1061 tel: (301) 496-2513; fax (301) 402-2389 email: [email protected]consensus, we do not answer the debate on which is the best method. Rather, we describe the technical issues and provide insights into the various performance tradeoffs. To maintain a focus on first pass perfusion methods, only certain topics could be included in this review. We start with the basic design for T1-weighted first pass myocardial perfu- sion imaging, and go on to discuss the tradeoffs associated with various schemes to provide multi-slice coverage. Artifact mech- anisms are discussed and related to the sequence design and parameters. For simplicity, we consider only gadolinium based extracellular contrast agents, although there may be advantages to alternative contrast agents. Both qualitative and quantitative analysis is considered in de- termining the various performance requirements, such as spatial resolution, temporal resolution, and T1-linearity. The topic of image analysis is another source of debate, and there remain questions on the value and accuracy of various performance in- dices. For issues beyond those dealt with in the current review, there is an abundance of literature on myocardial perfusion imag- ing, including original research on imaging methodology (3–17), comparisons between methods (18–22), review papers (23–27), and papers dealing with the subject of quantitative perfusion analysis (26, 28–33). MYOCARDIAL PERFUSION IMAGING Myocardial perfusion imaging is based on measuring the de- livery of contrast agent to the myocardium during the first pass following a bolus injection. The signal intensity is enhanced 525
Imaging Sequences for First PassPerfusion—A Review
Peter Kellman, PhD and Andrew E. Arai, MD
Laboratory of Cardiac Energetics, National Heart, Lung and Blood Institute,National Institutes of Health, Department of Health and Human Services, Bethesda, Maryland, USA
ABSTRACT
Myocardial perfusion imaging sequences and analysis techniques continue to improve. Wereview the state-of-the-art in cardiovascular magnetic resonance first pass perfusion pulse se-quences including the application of parallel imaging. There are a wide range of sequencedesigns and parameters to consider when optimizing an acquisition protocol. The interdepen-dence of these parameters forces the user to make compromises. We describe the technicalissues and provide insights into the various performance tradeoffs. We also review the basic de-sign for T1-weighted first pass myocardial perfusion imaging and go on to discuss the tradeoffsassociated with various schemes to provide multi-slice coverage. Artifact mechanisms are dis-cussed and related to sequence design and parameters. The selection of quantitative versusqualitative analysis affects various performance requirements, such as spatial and temporalresolution and linearity of enhancement. Understanding the interaction between the pulse se-quence parameters and resulting image quality is important for improving myocardial perfusionimaging.
INTRODUCTION
Myocardial perfusion imaging using cardiovascular magneticresonance (CMR) has undergone steady improvement since itwas first proposed over 20 years ago (1, 2). Yet there remainchallenges for widespread clinical acceptance, and there is stillconsiderable debate as to which technique is best. We review thestate-of-the-art first pass perfusion pulse sequences including theapplication of parallel imaging for accelerated acquisition. Per-formance optimization is difficult due to the large number ofparameters affecting image acquisition. Furthermore, the inter-dependence of these parameters forces the user to make com-promises. As the field has not come to clear methodological
Keywords: Myocardial Perfusion, Cardiovascular MR, Ischemia,Parallel Imaging.Supported by a grant from the National Heart, Lung, and BloodInstitute, National Institutes of Health, Intramural ResearchProgramCorrespondence to:Peter KellmanLaboratory of Cardiac EnergeticsNational Institutes of HealthNational Heart, Lung and Blood Institute10 Center Drive, MSC-1061Building 10, Room B1D416Bethesda, MD 20892-1061tel: (301) 496-2513; fax (301) 402-2389email: [email protected]
consensus, we do not answer the debate on which is the bestmethod. Rather, we describe the technical issues and provideinsights into the various performance tradeoffs.
To maintain a focus on first pass perfusion methods, onlycertain topics could be included in this review. We start withthe basic design for T1-weighted first pass myocardial perfu-sion imaging, and go on to discuss the tradeoffs associated withvarious schemes to provide multi-slice coverage. Artifact mech-anisms are discussed and related to the sequence design andparameters. For simplicity, we consider only gadolinium basedextracellular contrast agents, although there may be advantagesto alternative contrast agents.
Both qualitative and quantitative analysis is considered in de-termining the various performance requirements, such as spatialresolution, temporal resolution, and T1-linearity. The topic ofimage analysis is another source of debate, and there remainquestions on the value and accuracy of various performance in-dices. For issues beyond those dealt with in the current review,there is an abundance of literature on myocardial perfusion imag-ing, including original research on imaging methodology (3–17),comparisons between methods (18–22), review papers (23–27),and papers dealing with the subject of quantitative perfusionanalysis (26, 28–33).
MYOCARDIAL PERFUSION IMAGING
Myocardial perfusion imaging is based on measuring the de-livery of contrast agent to the myocardium during the first passfollowing a bolus injection. The signal intensity is enhanced
525
Dow
nloa
ded
By: [
LCM
R -
Jour
nal o
f Car
diov
ascu
lar M
agne
tic R
eson
ance
] At:
23:3
1 29
Mar
ch 2
007
by the contrast agent, which shortens the T1 relaxation timeand results in a brighter signal using a T1-weighted imagingsequence. Regions with lower regional blood flow will appearhypointense and may be detected given adequate image quality.Quantitative measurement of blood flow may be made throughanalysis of the dynamics of the myocardial signal intensity mea-surement as a function of time. Myocardial flow reserve may beestimated by comparing the flow measurements acquired at restand at stress. Stress perfusion is most commonly studied usingvasodilation such as adenosine or dipyridamole. Vasodilatorsincrease the blood flow in normal vessels while stenotic vesselshave a reduced vasodilator response.
Regions with prior myocardial infarction may appear hypo-intense despite normal blood flow, following revascularization,due to the low flow into scar tissue. Therefore, the interpreta-tion of perfusion images usually also incorporates viability as-sessment by delayed enhancement imaging as described in thesection on analysis.
Imaging requirements
Successful myocardial perfusion imaging requires optimiz-ing sequence and parameters to meet often contradictory re-quirements. The basic requirements are:
1. Temporal Resolution Two distinct measures of tempo-ral resolution are important for perfusion imaging. Thetime between two images of the same slice location af-fects the ability to sample the dynamic signal intensitychanges during the first pass to allow modeling the kinet-ics of blood flow to the myocardium. Typically, imagesare acquired every 1–2 heartbeats to adequately sam-ple myocardial blood flow. For quantitative perfusion,an accurate estimate of the arterial input function mayrequire sampling the LV blood signal every heartbeat.Also of importance is the time per slice (Tslice) within thecardiac cycle, and the actual duration of imaging read-out (Timaging), which determines the sensitivity to cardiacmotion, both indicated in Fig. 1a.
2. Spatial Resolution The spatial resolution must be ade-quate to distinguish sub-endocardial ischemia (<3 mmin-plane) and to assess transmural extent of defects.
3. Spatial Coverage It is desirable to have full coverage ofthe heart. A minimum of 3 slices is needed to cover atleast 16 segments of the heart (34). A greater number ofslices are desirable.
4. Linearity A linear or quantifiable relationship betweensignal intensity and contrast agent concentration is de-sirable in order to quantify perfusion.
5. Image Quality Image quality must be sufficient to pro-vide contrast between normal and ischemic regions andmust be free of artifacts.
A desire to quantify myocardial perfusion imposes additionalrequirements regarding accurate knowledge of the arterial inputfunction, which represents the delivery of contrast to the heartand is commonly estimated from the blood signal. Therefore,
quantitative myocardial perfusion imaging requires measuringboth the blood and myocardial signals. The blood and my-ocardium have different contrast agent concentrations as well asT1, T2, and T2∗ relaxation parameters leading to significantlydifferent imaging characteristics and signal intensities. The con-trast between blood and myocardium is also affected by bloodflow.
T1-weighted imaging sequences
First-pass perfusion imaging typically acquires multipleslices of T1-weighted images that portray perfusion. Ideally,the signal intensity on such images should closely reflect thetemporal change of the contrast agent concentration, which isinversely related to the achieved T1 ([Gd] ∼ 1/T1). A saturationrecovery (SR) preparation (Fig. 1a) is the most commonly im-plemented method to achieve T1-weighting and may be used inconjunction with various methods for image readout.
In early work, inversion recovery (IR) approaches (Fig. 1b)were used (7–9, 17, 35, 36). IR approaches have the potentialfor increased dynamic range but are vulnerable to R-R variationor missed triggers which will cause signal intensity variationdue to incomplete magnetization recovery. With IR preparation,the images are usually acquired at a time following inversion(TI) which nulls the pre-contrast blood to maximize the contrastand to avoid loss of contrast, which might result from mag-nitude detection. This results in a comparatively long imagingduration and commonly limits the imaging to a single slice perheartbeat.
526 P. Kellman and A. E. Arai
Dow
nloa
ded
By: [
LCM
R -
Jour
nal o
f Car
diov
ascu
lar M
agne
tic R
eson
ance
] At:
23:3
1 29
Mar
ch 2
007
In both SR and IR preparation schemes, there is generally atrigger delay (TD) between the RF preparation and the start ofthe image acquisition. In most cases, the signal intensity is deter-mined by the delay to the center of k-space, which is commonlyin the center of readout for linear phase encode ordering. Thistime is often referred to as the inversion time (TI) even in cases ofSR preparations. While TD and TI are closely related, the litera-ture contains reference to both often making direct comparisonsdifficult.
An alternative magnetization driven steady state preparationapproach (Fig. 1c) was proposed (10) to achieve a higher degreeof linearity than IR (i.e., signal intensity vs [Gd]). However,linearity was achieved at the expense of contrast-to-noise ratio(CNR) and had a lengthy preparation time limiting the acqui-sition to a single slice per heartbeat. In this scheme, the longi-tudinal magnetization was driven to steady state by a series ofRF (alpha) pulses, followed by the readout. The preparation usednon-selective RF pulses to reduce the effects of flow and motion.In the steady state limit, the signal intensity is linearly propor-tional to 1/T1 when using 90◦ readout pulses and was found tobe quite linear using 45◦ readout, providing increased sensitivity(improved SNR) with only slight compromise in linearity (10).
Hybrid T1-preparation schemes
A number of hybrid preparation schemes can achieve T1-weighting with specific advantages and disadvantages. Arrhyth-mia insensitive approaches which combine SR preparation withIR have been proposed (Fig. 2a) (37). This scheme increasesthe length of preparation thereby reducing the number of slicesthat may be covered. By using a 90◦ SR preparation, which setsthe longitudinal magnetization to zero, the 180◦ IR preparationwhich follows after a fixed short recovery period becomes in-sensitive to the RR interval. The dynamic range is somewhatcompromised depending on the saturation recovery delay.
The magnetization driven steady state scheme may also beused with a 90◦ SR preparation (Fig. 2b) to reduce the time
Figure 2. Hybrid magnetization preparation schemes for T1-weighted myocardial perfusion imaging: (a) arrhythmia insensitivepreparation using SR followed by IR, and (b) magnetization drivensteady state with SR preparation to decrease time to steady state.
required to reach steady state. However, the linearity is stillachieved at the expense of contrast-to-noise ratio (CNR) andslightly longer time per slice.
Multi-slice approaches
There are several approaches to providing multi-slice cover-age during first-pass imaging as shown in Fig. 3 for SR prepara-tion. Other preparation schemes might be implemented as wellif the time per slice permits. In each of these schemes, imagingis performed across the cardiac cycle; therefore, each slice willbe acquired at a different phase of the cardiac cycle. Thus, thewall thickness and motion will vary slice-to-slice using theseapproaches. While it is possible to acquire fewer slices with lessvariation, for instance, only during diastole, this may reduce theoverall spatial coverage, particularly if single RR temporal res-olution is required. Furthermore, diastolic images have reducedwall thickness placing a higher demand on spatial resolution.
Using a single SR preparation per slice (Fig. 1a) providesreasonable coverage with uniform image quality. Approachesthat acquire multiple slices per SR preparation offer increasedefficiency. In one approach, multiple slices are readout sequen-tially following a single shared SR preparation (Fig.3b) (38, 39).In this scheme the TI varies from slice-to-slice and thus the T1-weighted contrast, CNR, and linearity all vary slice to slice. Thisscheme does reduce the time per slice slightly providing greaterspatial coverage. A second approach which acquires multiple
Figure 3. Approaches to multi-slice coverage. (a) Multiple SRpreps per RR, (b) multiple slices readout sequentially per SR prep,(c) multiple interleaved slices per SR prep, and (d) multiple sliceselective notched pulse preparations per RR.
First Pass Perfusion Imaging 527
Dow
nloa
ded
By: [
LCM
R -
Jour
nal o
f Car
diov
ascu
lar M
agne
tic R
eson
ance
] At:
23:3
1 29
Mar
ch 2
007
Figure 4. Saturation performance comparison of different SR prep pulses for GRE-EPI perfusion sequence with TI = 85 ms. (a) proton densityreference without SR prep, (b) 90◦ SR prep pulse demonstrating B1 inhomogeneity and incomplete saturation, (c) sequel of 3 each 90◦ RECTpulses, (d) BIR4 adiabatic SR prep. All images are window-leveled the same. Incomplete saturation is evident using the 90◦ RECT (b).
slices per SR preparation acquires multiple interleaved slices(e.g., 2 slices) (Fig. 3c) (11). In this scheme, the TI remainsconstant. Although this approach increases the acquisition ef-ficiency and thus spatial coverage, the actual imaging time perindividual slice is increased thereby increasing the sensitivityto motion artifacts. A benefit of slice interleaving is the length-ening of the pulse repetition time (TR) between pulses of thesame slice. An increased TR permits greater magnetization re-covery and use of a higher readout flip angle for improvedSNR.
A novel scheme for multi-slice coverage uses selective notchpulse preparations (15) where each notch selective SR prepara-tion is used to prepare the 2nd slice following the saturation pulse(Fig. 3d). In this scheme, the selective notch pulse saturates thevolume except for the next slice to be imaged. Thus, the satura-tion preparation for any given slice was played out prior to theprevious slice readout. In this way, the preparation time (TI) maybe made longer without requiring the overhead of a trigger delay.A longer TI may result in increased SNR although this comes atthe cost of reduced linearity. A problem with this scheme is thatthe first slice may need to be discarded due to RR variations,eroding some of the efficiency gain. Furthermore, the selectivenotch pulse frequently results in non-uniform blood pool signalintensity due to heterogeneous through plane motion. Myocar-dial signal intensity is also sensitive to through plane motion,either cardiac or respiratory related.
Saturation recovery preparation
The 90◦ SR preparation is commonly used since an ideal90◦ SR is insensitive to arrhythmia and/or missed ECG triggers.SR preparations using a lower flip angle (60◦–70◦) have beenreported (3, 5) in cases of very short TD (10–15 ms) increasingsensitivity to RR variation. The 90◦ SR provides improved CNRgiven an adequate TI (3).
A few RF pulse designs have been proposed to achievethe ideal 90◦ SR preparation in the presence of B1-field in-homogeneity. The adiabatic BIR4 pulse provides improvedinsensitivity to both B1 and B0 (40, 41). The BIR4 has excel-
lent performance as long as the adiabatic condition is achieved(i.e., adequate power). Alternatively, a pulse sequel design withseveral repeated 90◦ rectangular pulses with gradient crushers(42) improves the overall effectiveness of the saturation pulsein the event that the 90◦ rectangular pulse is not a true 90◦ dueto B1-field inhomogeneity. In our experience both of these ap-proaches are very effective and make a significant improvementover the simple 90◦ rectangular pulse (Fig. 4). The sensitivity toRR variation is illustrated in Fig. 5, which shows the effect ofmissed ECG triggers on the signal intensity for both 90◦ rectan-gular SR preparation and pulse sequel (3 pulses) with imperfect90◦ pulses. The effect of missed ECG triggers is insignificantfor the pulse sequel design.
The gradient crushers used in conjunction with the SR prepa-ration RF pulse can lead to stimulated echo artifacts which de-grade image quality. Stimulated echo artifacts due to these gra-dients may be mitigated by use of variable crushers.
Figure 5. Time intensity curves for 2 cases with missed ECG trig-gers showing (a) sensitivity of signal intensity to missed triggersusing 90 RECT SR prep with incomplete saturation (left) and (b) in-sensitivity to missed triggers using pulse sequel for SR prep (right).Incomplete saturation causes increased intensity following missedtrigger (left). Samples following missed ECG trigger are markedwith circles.
528 P. Kellman and A. E. Arai
Dow
nloa
ded
By: [
LCM
R -
Jour
nal o
f Car
diov
ascu
lar M
agne
tic R
eson
ance
] At:
23:3
1 29
Mar
ch 2
007
The saturation performance of the 90◦ SR preparation is par-ticularly important for semi- and fully-quantitative perfusionmeasurements which rely on accurate baseline correction. Im-perfect saturation biases the pre-contrast baseline by an amountthat is not readily correctable.
Image readout & parallel imaging
Myocardial perfusion imaging sequences have been develop-ed using IR or SR recovery in combination with either snap-shot FLASH (7, 8), GRE-EPI (5), or balanced steady state freeprecession (SSFP, also known as FISP, FIESTA, or BalancedFFE) (13). There is no clear consensus regarding the sequenceof choice for myocardial perfusion imaging, although there isconsiderable debate and several published comparisons (18–22).
There are also a number of approaches to parallel imaging(43–47) and other accelerated imaging methods (48–50) thatmay be applied to myocardial perfusion imaging. The appli-cation of parallel imaging is closely coupled with the imagereadout since the number of phase encodes may be substan-tially reduced, thus, altering the optimization of sequence pa-rameters normally associated with the readout. The optimumchoice of methods is a difficult question due to the large para-meter space for each method and differing details of implemen-tation between institutions and vendor platforms. It is not theobjective of this paper to answer this debate. Rather, it is to re-view the current methods in light of advances, such as parallelimaging, and to highlight the pros and cons and performanceissues.
Tradeoffs between spatial resolution and SNR, spatial andtemporal resolution, and so on, are quite familiar in CMR. Withthe addition of the SR preparation, the number of variables andtheir interdependence grows. For example, a longer TI may in-crease the SNR but reduce the linearity of signal intensity vs con-trast agent concentration (see below section). Some sequenceshave been optimized for linearity (10) while others have beenoptimized for CNR (3). In general, a longer TI will also re-duce the spatial coverage. While one can optimize sequenceparameters for SNR, coverage, and resolution, the interplay ofsequence parameters and image artifacts is less understood. Asdescribed later, the presence of image artifacts in myocardialperfusion imaging is a large factor limiting clinical acceptance.The choice of readout and sequence parameters affects the pointspread function (PSF) and other artifact mechanisms. The PSFis the term used to describe imperfect mapping between pointson the object to the image, which may cause blurring or artifactsin the image.
In order to illustrate the influence of sequence parameters, anexample is presented comparing SSFP, FLASH, and GRE-EPIreadouts, all accelerated with SENSE parallel imaging at rateR = 2 (43, 44). The sequence parameters (Table 1) are based onvalues from current state-of-the-art implementations reported inthe recent literature (21, 51) and are based on high performancegradients capable of 45 mT/m at 200 mT/m/ms slew rate. Thisexample serves to illustrate the relative SNR and CNR, linearity,and the uniformity of k-space which affects the PSF.
In these examples, the matrix size = 128 × 80, and the TI =85 ms. Both are held constant between sequences. All the se-quences provide similar spatial coverage (slices per RR), withthe GRE-EPI providing slightly greater coverage at the highestheart rates due to the increase acquisition efficiency of an EPIreadout (Tslice = 117 ms vs 130 or 132 ms). The most signif-icant difference given these parameters is the actual imagingduration within the cardiac cycle (Timaging), which ranges from61 ms for GRE-EPI to 92 ms for SSFP. The imaging durationmay contribute to motion induced artifacts especially near theendocardium. Selection of the bandwidth (BW) is a tradeoff be-tween the SNR and minimizing TR which affects Timaging, Tslice,and number of slices per RR. It also affects the TE which in turndetermines the T2∗ loss at peak contrast concentrations and thesensitivity to off-resonance due to B0-inhomogeneity and bolussusceptibility. TE is particularly important for a SSFP readout.While CNR is an important factor that depends on the
√BW, the
reduction of artifacts due to motion sensitivity (proportional toTimaging) was judged to be of paramount importance. The TR per-iod which consists of the RF pulse and actual readout decreasedwith increasing BW, reaching a point where the overhead of theRF pulse duration is a significant fraction. At this point, thereis diminishing return in further BW increase. In this example(Table 1), the BWs were selected accordingly. Using a constantreadout flip angle, the value of flip angle was selected for bestuniformity of k-space response. Alternatively, a ramped or othervariable readout flip angle might be considered.
The simulated responses of myocardial signal intensity areshown in Fig. 6 for varying contrast agent concentrations from0 to 4 mmol/L, where the expected concentration in the my-ocardium is in the range 1–2 mmol/L for single dose (0.1mmol/kg). Simulations used the method by Sekihara (52) withprecontrast T1 = 850 ms and T2 = 50 ms in the myocardium,and 4.5 (sec mmol/L)−1 relaxivity of Gd based contrast agent.Note that the 40 actually acquired phase encodes are recon-structed to a full resolution of 80 lines by means of paral-lel imaging. Non-uniformity of the k-space response due toSR recovery and transient approach to the steady state of thereadout leads to distortion of the PSF. Distorted PSF maycause edge artifacts or ghosting. The GRE-EPI has been re-ordered based on a modified center-out phase encode order (5)
First Pass Perfusion Imaging 529
Dow
nloa
ded
By: [
LCM
R -
Jour
nal o
f Car
diov
ascu
lar M
agne
tic R
eson
ance
] At:
23:3
1 29
Mar
ch 2
007
Figure 6. Simulated magnetization for SR myocardial perfusion imaging sequences with SSFP (left), FLASH (center), and GRE-EPI (right)readouts (parameters in Table 1) for contrast agent concentration ranging from 0 to 4 mmol/L (in 0.5 steps). Note that these plots representmagnetization signal and must be scaled by
√(BW) to compare relative SNR. The actual number of acquired phase encodes = 40 is reconstructed
to 80 using R = 2 parallel imaging.
chosen to acquire the central portion of k-space on the firstecho (TE1) of the echotrain, thereby minimizing T2∗ lossesand flow sensitivity. This results in a rapid periodic fluctuationin k-space which may lead to ghosting if the amplitude is toolarge.
The SSFP readout has the greatest signal (transverse mag-netization), while the FLASH readout has the least signal. TheSNR and CNR must also factor in the bandwidths (see followingsection).
The imaging parameters may be selected to meet differentcriteria. Note that the TI for each sequence might be variedto effect greater uniformity across k-space or variable readoutflip angle might be considered. It may be noted that the SSFPsequence achieves higher SNR than GRE-EPI for instance, butrequires a longer acquisition time. The SSFP sequence mightbe accelerated at R = 3 reducing Timaging from 92 ms to 61 ms(same as GRE-EPI) trading SNR for speed, making the SSFPand GRE-EPI readout methods practically equivalent in imagingduration and SNR. The SNR decrease between R = 3 and 2 is√
(3/2) considering the SNR loss of√
R.
Figure 7. Spatial coverage and spatial resolution versus heart rate for SSFP, FLASH, and GRE-EPI sequences with and without parallel imaging.R = 2 denotes parallel imaging at acceleration rate 2, and R = 1 denotes no acceleration.
IMAGING PERFORMANCE
Spatial resolution & coverage
The tradeoff between spatial resolution and coverage for SRwith various readouts is described in Fig. 7, which plots thenumber of slices acquired per RR at heart rates of 60, 90, and 120bpm for various matrix sizes and parallel imaging accelerationrates. Using a 360 × 270 mm2 rectangular FOV, the spatialresolution is 2.8 × 3.4 mm2 using a 128 × 80 matrix size, and1.9 × 2.8 mm2 using a 192 × 96 matrix size. These calculationsassume TI = 100 ms for 128 × 80 at rate = 1 (no acceleration),TI=85 ms for 128×80 at rate 2, TI=150 ms for 192×96 at rate1, and TI = 125 ms for 192 × 96 at rate 2. Using a 192 readoutresolution extends the TR from 2.3 to 2.6 ms for SSFP, 2.2 to 2.3for FLASH, and 6.1 to 6.6 for GRE-EPI. All sequence methodsmay achieve 3 slices per RR coverage at 192 × 96 matrix sizeat 120 bpm using parallel imaging at rate = 2. Of course, thehigher spatial resolution will decrease the SNR proportional tothe voxel size. In our experience, vasodilated stress heart rates areconsiderably higher than rest. Thus, pulse sequence parameters
530 P. Kellman and A. E. Arai
Dow
nloa
ded
By: [
LCM
R -
Jour
nal o
f Car
diov
ascu
lar M
agne
tic R
eson
ance
] At:
23:3
1 29
Mar
ch 2
007
Figure 8. Relative CNR versus dose for SSFP, GRE-EPI, and FLASH readouts after baseline correction and scaling for bandwidth (sequenceparameters per Table 1).
should be selected to achieve adequate spatial coverage at stressheart rates.
CNR & signal intensity linearity
The exponential recovery following saturation results in anon-linear relationship between signal intensity and contrastagent concentration. Ignoring the effect of readout on magneti-zation, the magnetization recovery following saturation is sim-ply described by M = [1- exp(-TI/T1)], with T1 described by(1/T1) = (1/T1o)+ (γ [Gd]), where T1o = 850 ms is the pre-contrast T1, and γ = 4.5 (sec mmol/L)−1 is the relaxivity ofGd based contrast agent with concentration [Gd]. For TI<<T1,the magnetization is approximately proportional to [Gd] as seenby simple substitution, since exp(-TI/T1) ≈ (1 − TI/T1). In theexample (Table 1) with TI = 85 ms, there is already signifi-cant departure from linearity at [Gd] = 1.5 mmol/L where T1 ≈125 ms.
A more comprehensive simulation that accounts for magneti-zation effects of readout yields curves (Fig. 8) that demonstratethe non-linear relation versus contrast dose for the example se-quence parameters listed in Table 1. The upper plots are thetransverse magnetization without scaling for the bandwidth usedin each sequence. The lower plots correspond to relative CNRafter scaling for
√BW and correction of baseline intensity. The
CNR for these plots is defined as signal difference between pre-contrast and post-contrast at a myocardial concentration of 2mmol/L divided by the noise standard deviation. The CNR ver-sus TI (Fig. 9) shows that small increases in CNR are possiblewith a large increase in non-linearity. Optimization for CNRwithout consideration of linearity and k-space uniformity willlead to selection of long TI (3).
The simulation for the parameters of Table 1 predicts thatSSFP has approximately 40% higher CNR than GRE-EPI and
80% higher than FLASH, and GRE-EPI has approximately 40%higher CNR than FLASH. These predictions are fairly consis-tent with reported measurements using similar parameters (21)and with results that use FLASH with GRAPPA (45) parallelimaging (51) when factoring the effective acceleration used. Forexample (51), using R = 2 GRAPPA with 96 phase encode linesand 12 extra in-place reference lines results in an effective ac-celeration of R = 1.6.
The magnetization driven steady state approach (10) achievesa high degree of linearity even at 45◦ readout flip angle; however,the CNR penalty is estimated to be 50% or greater compared toSR-FLASH.
Artifacts
The presence of image artifacts in myocardial perfusionimaging is a large factor limiting clinical acceptance. In partic-ular, artifacts that appear as a dark subendocardial rim (example
Figure 9. Relative CNR versus TI (between 0 and 2 mmol/L) forSSFP, GRE-EPI, and FLASH readouts after baseline correctionand scaling for bandwidth (other parameters per Table 1).
First Pass Perfusion Imaging 531
Dow
nloa
ded
By: [
LCM
R -
Jour
nal o
f Car
diov
ascu
lar M
agne
tic R
eson
ance
] At:
23:3
1 29
Mar
ch 2
007
Figure 10. Dark rim artifact observed on both stress (left) and rest (right) studies using SR-SSFP for patient with negative cath (images courtesyof Jonathan Lyne, Royal Brompton Hospital, London).
Fig. 10) that may be confused with actual hypointense regionsof reduced blood flow (53). Such artifacts can lead to false diag-noses and are of paramount concern. Strategies (54) have beenarticulated to help discriminate these artifacts (see next section).Understanding the artifact mechanisms may lead to sequencedesigns which minimize these deleterious effects.
Artifact mechanisms that may lead to dark sub-endocardialrim artifacts include cardiac motion (55) during the actual imag-ing period (Timage), Gibb’s ringing (53) caused by truncation ofk-space, non-uniformity of k-space weighting due to saturationrecovery and readout, and partial volume cancellation betweenthe myocardium and LV blood pool. Contribution of each ofthese artifact mechanisms is affected by the sequence and choiceof parameters.
Minimizing cardiac motion artifacts may be accomplishedeither by using a small imaging duration (Timaging) or by timingthe acquisition to occur in periods with relatively low motion.Reducing the imaging duration may be accomplished by re-ducing the matrix size (low spatial resolution), using acceler-ated imaging such as parallel imaging, and using short TRsand/or EPI sequences. Diastole may provide a longer motionfree window but comes at the cost of reduced left ventricularwall thickness thereby increasing the demand for better spatialresolution.
Gibb’s ringing is generally mitigated to some extent by re-construction with the use of raw filtering, also known as win-dowing or apodization. The ringing is suppressed at the expenseof spatial resolution. In this case, it may be advantageous toacquire a larger matrix size and use a stronger raw filter (53).This will cost some spatial coverage but might be a worthwhiletradeoff. The extent of Gibb’s ringing is determined by the stepin intensity or contrast between the myocardium and blood.The SSFP sequence has a much higher blood-myocardiumcontrast, and therefore, will have commensurately largerringing.
The non-uniformity across k-space (Fig. 6) leads to pointspread function (PSF) distortion causing both blurring (loss ofspatial resolution) and edge enhancement, which can cause darkrim artifacts. With the exception of the magnetization driven
steady state (10) approach (Fig. 1[c]), myocardial perfusionimaging is not performed in a steady state condition, and thenon-uniformity of the k-space response is determined by the tis-sue T1 values, TI, and readout flip angle. The dependence on T1means that the PSF for the blood pool during peak bolus con-centrations may be quite different than for the myocardium andmay lead to rim artifacts at the sub-endocardial border betweenblood and myocardium. Ramp flip angle techniques might beconsidered to improve uniformity but are only effective for asingle T1 value. In EPI sequences, which acquire at multipleecho times, the T2* loss at peak bolus concentration contributesto significant non-uniformity of k-space weighting for the bloodpool signal. Non-uniform k-space weighting with the EPI phaseencode order may lead to ghosting.
Partial volume effects may lead to dark rim artifacts whenthe blood and myocardium are out of phase. Spatial variationin phase may result from a number of effects including stronggradients in contrast agent concentration. These effects are re-lated to the spatial resolution, therefore, it is desirable to haveas many pixels transmurally as possible.
Other artifacts occur in addition to the dark rim artifact, al-though these may have less serious consequences. Using a SSFPreadout, dark banding artifacts arise due to B0-field inhomo-geneity caused by inadequate shim or susceptibility gradientassociated with the bolus of contrast agent. Shim problems canbe observed prior to contrast injection and mitigated by centerfrequency or shim re-adjustment. GRE-EPI readout is marredby ghosting due to off-resonance since there are phase shiftsat each echo delay. This can be mitigated to a large extent byusing an interleaved phase encode order with echo-time shifting(5). Chemical shift due to fat may cause ghosting; therefore,fat suppression pulses (56) are often used in conjunction with aGRE-EPI readout and can be implemented during the prepara-tion time with little to no additional overhead. Stimulated echoesfrom the SR preparation may cause ghosting. Stimulated echoesmay be eliminated by means of time varying gradient spoiling.
Aliasing artifacts or wrap due to accelerated imaging maylead to interference in the region of interest (Fig. 11). Whilethese artifacts are generally recognized and therefore do not have
532 P. Kellman and A. E. Arai
Dow
nloa
ded
By: [
LCM
R -
Jour
nal o
f Car
diov
ascu
lar M
agne
tic R
eson
ance
] At:
23:3
1 29
Mar
ch 2
007
Figure 11. Examples images illustrating residual artifacts in UNFOLD reconstruction due to dynamic signal fluctuation. Case 1 illustrates UNFOLDartifact due to dynamic contrast enhancement of RV. Case 2 illustrates UNFOLD artifact due to breathing motion. In this example, these artifactsare suppressed with TSENSE parallel imaging.
the same consequence as the rim artifact, they may degrade thetime intensity curves preventing accurate analysis. Since theseartifacts may not appear until contrast is delivered, they can leadto non-diagnostic exams.
As illustrated in this section on artifacts, there are a widerange of factors that can result in artifacts. The worst artifactsresult in subendocardial rims that can be misdiagnosed as per-fusion defects. Factors that mitigate one mechanism generatingrim artifacts may worsen another factor. For example, Gibbsringing and partial volume errors can be improved by increas-ing image resolution, but higher spatial resolution may requirelonger readouts that exacerbate motion artifacts. Recognizingthat some artifacts are worse during the time when concentrationof gadolinium is highest in the ventricular cavities is importantsince this is a time when real perfusion defects should also bepresent.
ANALYSIS
Qualification
Qualitative assessment of myocardial perfusion deficit israpid and has achieved diagnostic levels of sensitivity and speci-ficity (54, 57–61) while quantitative analysis is currently moretime consuming. A qualitative readout basically consists of ex-amining the time course of images for evidence of hypointenseregions.
Due to issues of noise and artifacts, additional interpreta-tion strategies have been developed to minimize false positivediagnoses due to dark subendocardial rim artifacts. One strat-egy tested by the group at Duke University (54) combines stressand rest perfusion studies with delayed enhancement accordingto the following logic. Interpretation of coronary artery disease
(CAD) begins with delayed enhancement. Positive delayed en-hancement is a highly specific indicator of CAD. Negative de-layed enhancements leads to examining the stress perfusionstudy. Negative delayed enhancement combined with negativestress perfusion results in a negative diagnosis for CAD. How-ever, a positive stress perfusion deficit requires an analysis ofrest perfusion. If the rest perfusion indicated that the hypointenseregion coincides with the stress perfusion hypointense region,then the result is qualified as a rim artifact, under the suppo-sition that the rest study should have normal flow in regionswithout prior MI. Finally, an apparent stress perfusion deficitwith normal rest perfusion would indicate CAD as illustrated inexample of Fig. 12, which had no delayed enhancement. Thisstrategy deals effectively with no prior recognized MI but doesnot address cases of ischemia in patients with prior MI. An ex-ample of stress and rest studies with a dark rim artifact is shownin Fig. 10, which was corroborated by a negative finding on adiagnostic catheterization (example provided by Jonathan Lyneand Peter Gatehouse, Royal Brompton Hospital).
The simple strategy outlined above may be further augmentedby qualitative analysis of time intensity curves. This procedurerequired tracing endo- and epi-cardial contours to divide themyocardium into sectors. Each sector may be further subdi-vided transmurally into endo and epi layers. Tracing contourscan be time consuming when there is respiratory motion. Inthis case, automatic registration methods may be applied (62,63), although the performance of automatic registration is stillimproving.
Time intensity curves may be assessed qualitatively as wellas quantitatively. For stress perfusion, the expected response fornormal vessels with vasodilation show a peak (Fig. 13) in themyocardial signal intensity time course followed by a washout
First Pass Perfusion Imaging 533
Dow
nloa
ded
By: [
LCM
R -
Jour
nal o
f Car
diov
ascu
lar M
agne
tic R
eson
ance
] At:
23:3
1 29
Mar
ch 2
007
Figure 12. Example first-pass contrast-enhanced perfusion images for patient with stress perfusion deficit in antero-septal region shown forsingle slice of 3 acquired slices using GRE-EPI sequence using rate-2 TSENSE. The bottom row is at rest and top row is with stress: (a), (e)pre-contrast, (b), (f) RV enhanced, (c), (g) LV enhanced, and (d), (h) myocardium enhanced. Delayed enhancement images were negative forMI.
prior to the plateau. Absence of this overshoot indicates a lackof vasodilation and may be used to further augment the inter-pretation of findings based on visual image assessment.
Quantification
First-pass myocardial perfusion images may be used to char-acterize myocardial blood flow (MBF) using either fully or semi-quantitative methods. MBF (expressed in mL/min/g) and my-ocardial perfusion reserve (MPR) are defined as the ratio ofhyperemic and resting blood flow. Both are clinically importantindices for assessing myocardial ischemia, and are more objec-tive than qualitative assessment. Fully quantitative myocardialblood flow (MBF) may be estimated using a Fermi model con-strained deconvolution (29, 31, 32).
In order to quantify perfusion, it is necessary to have an ac-curate estimate of the arterial input function (AIF), which isnormally measured from the LV blood pool signal. Due to non-linear effects of saturation recovery and T2∗ losses at high bo-lus concentration, the AIF estimated directly from the myocar-dial perfusion images becomes significantly distorted. For thisreason, various solutions have been proposed. The dual-bolus
Figure 13. Time intensity curves for stress perfusion study (Fig. 12)using 6 endocardial sectors. Normal sectors show a vasodilatedresponse, whereas the antero-septal region is hypo-intense.
first-pass perfusion method uses a high dose of contrast for my-ocardial analysis, preceded by a lower concentration bolus tomaintain the linearity of the left ventricle (LV) signal intensity(29, 32). The dual sequence method (6, 12, 64), which acquiresAIF reference images using a low TE and short saturation re-covery delay, has been proposed to avoid distortion in the AIF.In the dual bolus method, a 1/20 dose mini-bolus has been foundto provide an adequately linear response without noticeable dis-tortion. The dual sequence method (Fig. 14) acquires low reso-lution blood pool images each heart beat requiring on the orderof 60 ms, which slightly reduces the maximum spatial coverage.In order to maintain linearity at peak bolus concentration, it isnecessary to have a TI on the order of 10 ms or less. This isaccomplished using a center-out phase encode acquisition or-der with a short trigger delay to avoid edge enhancement dueto highly non-uniform k-space response and a very low spatialresolution image. TE values on the order of 0.6 ms are used tominimize T2∗ distortion of the AIF (6, 64). T2∗ is estimated to bein the 6–12 ms range in the LV blood pool at peak concentrationsfor single dose (Fig. 15), corresponding to T2∗ losses of 5–10%using the dual sequence method with TE = 0.6 ms for the AIFmeasurement (65). The dual sequence method could be modi-fied for greater spatial coverage by sampling of the myocardialperfusion signal every other heart beat while still maintainingsingle heartbeat temporal resolution for the AIF.
The myocardial signal intensity does not have a linear rela-tionship to [Gd] as desired due to T1-related effects. Althoughthis has received less attention than the highly non-linear dis-tortion of the blood pool signal, the effect of T1-nonlinearityon myocardial signal intensity may significantly affect perfu-sion quantification (66, 67). The myocardial perfusion signalnon-linearity for SR sequences is largely determined by the TIas previously described and will affect the estimates of MBFeven for relatively short TI (67). Look-up table (LUT) correc-tion may be used to reduce the distortion providing more reliableflow estimates (66, 67). The LUT, which may be based on either
534 P. Kellman and A. E. Arai
Dow
nloa
ded
By: [
LCM
R -
Jour
nal o
f Car
diov
ascu
lar M
agne
tic R
eson
ance
] At:
23:3
1 29
Mar
ch 2
007
Figure 14. Dual sequence method for estimating arterial input function using a low resolution image with short TD for linear response of LVblood-pool signal at high Gd concentrations, and short TE to minimize T2∗ effects at peak bolus. Note that low spatial resolution AIF image(upper right) is adequate for sampling LV blood pool time intensity curve, but does not have adequate resolution to detect myocardial perfusiondeficit (lower right image).
Figure 15. T2∗ effects observed in time intensity curves for LVblood pool ROI measured at different TE values using a multi-echosequence. The TE = 0 curve (dotted line) is estimated based on aleast squares fit to the multi-echo dataset. The initial 2 time framesare proton density reference images.
Figure 16. Correction of inhomogeneity due to surface coil inten-sity variation performed by scaling the raw myocardial time inten-sity curves (left) for each myocardial sector ROI by the value ofthe initial acquired proton density reference image for the samecorresponding sector ROI (right).
theoretical or simulated curves, corrects the measured signalintensity for the non-linear relationship to Gd concentration.
Signal intensity variation due to surface coil B1-field inhomo-geneity will affect quantitative assessment and must be factoredinto LUT correction of non-linear signal intensity response. Sig-nal coil intensity correction may be implemented by normal-ization with proton density weighted images. Proton densityweighted images may be acquired either in a separate pre-scanor as part of the myocardial perfusion imaging sequence at theinitial frames just prior to the injection of contrast agent (using alow readout flip angle without SR preparation), ensuring imageregistration. Fig. 16 illustrates the time intensity curved beforeand after surface coil correction.
Typical acquisition of first-pass perfusion studies are 40-50heartbeats in duration, which is too long for a single breath-hold in many patients. Registration of the images with endoand epi-cardial borders is, therefore, a critical step in generatinghigh quality time intensity curves used in quantitative perfusionmeasurement (62, 63). Unless this step is reliably automated,this is the most time consuming aspect affecting the analysis.
A fully quantitative myocardial perfusion protocol generallyincludes both the stress and rest study in order to assess MPR andfor qualitative interpretation of dark rim artifacts as previouslydescribed. The stress study, which is most critical, is performedfirst. There will be a residual concentration of contrast agentfor the rest study, which is typically performed at least 10–15minutes following the first bolus.
It is important to note that the measure of Gd contrast agentis detected indirectly through its effect on the 1H signal. Thus,water exchange between the vascular space, the interstitial space,and the intercellular space must be considered (68).
CONCLUSIONS
Myocardial perfusion imaging sequences and analysis tech-niques continue to improve. There are a wide range of sequencedesigns and parameters to consider in order to optimize an acqui-sition protocol. There is need for consensus on both sequence and
First Pass Perfusion Imaging 535
Dow
nloa
ded
By: [
LCM
R -
Jour
nal o
f Car
diov
ascu
lar M
agne
tic R
eson
ance
] At:
23:3
1 29
Mar
ch 2
007
analysis in order to gain wider clinical acceptance and regulatoryapproval. There is also a need for common nomenclature. Darkrim artifacts may be minimized by careful design but remain asignificant limitation of myocardial perfusion imaging and mustbe dealt with by objective assessment such as quantitative mea-surement or careful qualitative interpretation schemes.
3. Bertschinger KM, Nanz D, Buechi M, Luescher TF, Marincek B, vonSchulthess GK, et al. Magnetic resonance myocardial first-passperfusion imaging: parameter optimization for signal response andcardiac coverage. J Magn Reson Imaging 2001;14:556–62.
4. Debatin JF, McKinnon GC, von Schulthess GK. Technical note–approach to myocardial perfusion with echo planar imaging.Magma 1996;4:7–11.
5. Ding S, Wolff SD, Epstein FH. Improved coverage in dynamiccontrast-enhanced cardiac MRI using interleaved gradient-echoEPI. Magn Reson Med 1998;39:514–9.
6. Gatehouse PD, Elkington AG, Ablitt NA, Yang GZ, Pennell DJ,Firmin DN. Accurate assessment of the arterial input function dur-ing high-dose myocardial perfusion cardiovascular magnetic res-onance. J Magn Reson Imaging 2004;20:39–45.
7. Haase A. Snapshot FLASH MRI. Applications to T1, T2, andchemical-shift imaging. Magn Reson Med 1990;13:77–89.
8. Haase A, Matthaei D, Bartkowski R, Duhmke E, Leibfritz D. In-version recovery snapshot FLASH MR imaging. J Comput AssistTomogr 1989;13:1036–40.
9. Jivan A, Horsfield MA, Moody AR, Cherryman GR. DynamicT1 measurement using snapshot-FLASH MRI. J Magn Reson1997;127:65–72.
10. Judd RM, Reeder SB, Atalar E, McVeigh ER, Zerhouni EA. Amagnetization-driven gradient echo pulse sequence for the studyof myocardial perfusion. Magn Reson Med 1995;34:276–82.
11. Kellman P, Derbyshire JA, Agyeman KO, McVeigh ER, AraiAE. Extended coverage first-pass perfusion imaging using slice-interleaved TSENSE. Magn Reson Med 2004;51:200–204.
12. Kim D, Axel L. Multislice, dual-imaging sequence for increasingthe dynamic range of the contrast-enhanced blood signal andCNR of myocardial enhancement at 3T. J Magn Reson Imaging2006;23:81–6.
13. Schreiber WG, Schmitt M, Kalden P, Mohrs OK, Kreitner KF, The-len M. Dynamic contrast-enhanced myocardial perfusion imag-ing using saturation-prepared TrueFISP. J Magn Reson Imaging2002;16:641–52.
14. Schwitter J, Debatin JF, von Schulthess GK, McKinnon GC. Normalmyocardial perfusion assessed with multishot echo-planar imag-ing. Magn Reson Med 1997;37:140–47.
16. Wendland MF, Saeed M, Masui T, Derugin N, Moseley ME, HigginsCB. Echo-planar MR imaging of normal and ischemic myocardiumwith gadodiamide injection. Radiology 1993;186:535–42.
17. Wendland MF, Saeed M, Yu KK, Roberts TP, Lauerma K, DeruginN, et al. Inversion recovery EPI of bolus transit in rat myocardiumusing intravascular and extravascular gadolinium-based MR con-trast media: dose effects on peak signal enhancement. Magn Re-son Med 1994;32:319–29.
19. Fenchel M, Helber U, Simonetti OP, Stauder NI, Kramer U, NguyenCN, et al. Multislice first-pass myocardial perfusion imaging: Com-parison of saturation recovery (SR)-TrueFISP-two-dimensional(2D) and SR-TurboFLASH-2D pulse sequences. J Magn ResonImaging 2004;19:555–63.
20. Wang Y, Moin K, Akinboboye O, Reichek N. Myocardial first passperfusion: steady-state free precession versus spoiled gradientecho and segmented echo planar imaging. Magn Reson Med2005;54:1123–9.
21. Weber S, Kronfeld A, Kunz RP, Fiebich M, Kreitner KF, SchreiberWG. Comparison of three accelerated Pulse Sequences for Quan-titative Myocardial Perfusion Imaging using TSENSE. Proc Intl SocMag Reson Med 2006;14:206.
23. Barkhausen J, Hunold P, Jochims M, Debatin JF. Imaging of my-ocardial perfusion with magnetic resonance. J Magn Reson Imag-ing 2004;19:750–57.
24. Cuocolo A, Acampa W, Imbriaco M, De Luca N, Iovino GL, Salva-tore M. The many ways to myocardial perfusion imaging. Q J NuclMed Mol Imaging 2005;49:4–18.
25. Wilke N. MR measurement of myocardial perfusion. Magma1998;6:147.
26. Wilke NM, Jerosch-Herold M, Zenovich A, Stillman AE. Magneticresonance first-pass myocardial perfusion imaging: clinical valida-tion and future applications. J Magn Reson Imaging 1999;10:676–85.
27. Canet EP, Janier MF, Revel D. Magnetic resonance perfusionimaging in ischemic heart disease. J Magn Reson Imaging1999;10:423–33.
28. Axel L. Tissue mean transit time from dynamic computed to-mography by a simple deconvolution technique. Invest Radiol1983;18:94–9.
29. Christian TF, Rettmann DW, Aletras AH, Liao SL, Taylor JL,Balaban RS, et al. Absolute myocardial perfusion in caninesmeasured by using dual-bolus first-pass MR imaging. Radiology2004;232:677–84.
30. Jerosch-Herold M, Seethamraju RT, Swingen CM, Wilke NM, Still-man AE. Analysis of myocardial perfusion MRI. J Magn ResonImaging 2004;19:758–70.
31. Jerosch-Herold M, Wilke N, Stillman AE. Magnetic reso-nance quantification of the myocardial perfusion reserve with aFermi function model for constrained deconvolution. Med Phys1998;25:73–84.
32. Hsu LY, Rhoads KL, Holly JE, Kellman P, Aletras AH, AraiAE. Quantitative myocardial perfusion analysis with a dual-boluscontrast-enhanced first-pass MRI technique in humans. J MagnReson Imaging 2006;23:315–22.
33. Kostler H, Ritter C, Lipp M, Beer M, Hahn D, Sandstede J. Pre-bolus quantitative MR heart perfusion imaging. Magn Reson Med2004;52:296–9.
34. Cerqueira MD, Weissman NJ, Dilsizian V, Jacobs AK, Kaul S,Laskey WK, et al. Standardized myocardial segmentation andnomenclature for tomographic imaging of the heart: a statementfor healthcare professionals from the Cardiac Imaging Committeeof the Council on Clinical Cardiology of the American Heart Asso-ciation. Circulation 2002;105:539–42.
35. Fritz-Hansen T, Rostrup E, Ring PB, Larsson HB. Quantification ofgadolinium-DTPA concentrations for different inversion times usingan IR-turbo flash pulse sequence: a study on optimizing multisliceperfusion imaging. Magn Reson Imaging 1998;16:893–9.
536 P. Kellman and A. E. Arai
Dow
nloa
ded
By: [
LCM
R -
Jour
nal o
f Car
diov
ascu
lar M
agne
tic R
eson
ance
] At:
23:3
1 29
Mar
ch 2
007
36. Wilke N, Simm C, Zhang J, Ellermann J, Ya X, Merkle H, etal. Contrast-enhanced first pass myocardial perfusion imaging:correlation between myocardial blood flow in dogs at rest and dur-ing hyperemia. Magn Reson Med 1993;29:485–97.
37. Tsekos NV, Zhang Y, Merkle H, Wilke N, Jerosch-Herold M,Stillman A, Ugurbil K. Fast anatomical imaging of the heartand assessment of myocardial perfusion with arrhythmia insen-sitive magnetization preparation. Magn Reson Med 1995;34:530–36.
38. Nagel E, Klein C, Paetsch I, Hettwer S, Schnackenburg B,Wegscheider K, et al. Magnetic resonance perfusion measure-ments for the noninvasive detection of coronary artery disease.Circulation 2003;108:432–7.
39. Fischer SE, Lorenz CH. [Determining heart muscle perfusion bymagnetic resonance tomography progressing to clinical applica-tion]. Radiologe 1997;37:366–71.
40. Kim D, Cernicanu A, Axel L. B(0) and B(1)-insensitive uniform T(1)-weighting for quantitative, first-pass myocardial perfusion magneticresonance imaging. Magn Reson Med 2005;54:1423–9.
41. Staewen RS, Johnson AJ, Ross BD, Parrish T, Merkle H, GarwoodM. 3-D FLASH imaging using a single surface coil and a new adi-abatic pulse, BIR-4. Invest Radiol 1990;25:559–67.
42. Oesingmann N, Zhang,Q. ,Simonetti, O. Improved Saturation RFPulse Design for Myocardial First Pass Perfusion at 3T. Journal ofCardiovascular Magnetic Resonance 2004;6:373–4.
43. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE:sensitivity encoding for fast MRI. Magn Reson Med 1999;42:952–62.
49. Madore B, Glover GH, Pelc NJ. Unaliasing by fourier-encodingthe overlaps using the temporal dimension (UNFOLD), appliedto cardiac imaging and fMRI. Magn Reson Med 1999;42:813–28.
50. Di Bella EV, Wu YJ, Alexander AL, Parker DL, Green D, Mc-Gann CJ. Comparison of temporal filtering methods for dynamiccontrast MRI myocardial perfusion studies. Magn Reson Med2003;49:895–902.
51. Lyne JC, Assomull, R, Smith, GC, Gatehouse PD, Firmin DN, Pen-nell DJ. Comparison of EPI, TrueFISP, and FLASH Sequences withParallel Acquisition for Myocardial Perfusion Imaging. Journal ofCardiovascular Magnetic Resonance 2006;8:113–4.
52. Sekihara K. Steady-state magnetizations in rapid NMR imagingusing small flip angles and short repetition intervals. IEEE Transon Med Imag 1987;6:157–64.
53. Di Bella EV, Parker DL, Sinusas AJ. On the dark rim artifact in dy-namic contrast-enhanced MRI myocardial perfusion studies. MagnReson Med 2005;54:1295–9.
54. Klem I, Heitner JF, Shah DJ, Sketch MH, Jr., Behar V, Weinsaft J,Cawley P, Parker M, Elliott M, Judd RM, Kim RJ. Improved detec-tion of coronary artery disease by stress perfusion cardiovascularmagnetic resonance with the use of delayed enhancement infarc-tion imaging. J Am Coll Cardiol 2006;47:1630–8.
55. Storey P, Chen Q, Li W, Edelman RR, Prasad PV. Band artifactsdue to bulk motion. Magn Reson Med 2002;48:1028–36.
56. Ingkanisorn WP, Rhoads KL, Syed, MA, Dyke, CK, Kellman P, Ale-tras AH, et al. Fat-Suppression Improves Image Quality and Diag-nostic Accuracy of EPI First Pass Perfusion. Proc of Intl Soc MagRes Med. 2004; Kyoto, Japan; 312.
57. Ingkanisorn WP, Kwong RY, Bohme NS, Geller NL, Rhoads KL,Dyke CK, et al. Prognosis of negative adenosine stress magneticresonance in patients presenting to an emergency department withchest pain. J Am Coll Cardiol 2006;47:1427–32.
58. Ishida N, Sakuma H, Motoyasu M, Okinaka T, Isaka N, Nakano T,et al. Noninfarcted myocardium: correlation between dynamic first-pass contrast-enhanced myocardial MR imaging and quantitativecoronary angiography. Radiology 2003;229:209–16.
59. Paetsch I, Foll D, Langreck H, Herkommer B, Klein C, Schalla S, etal. Myocardial perfusion imaging using OMNISCAN: a dose findingstudy for visual assessment of stress-induced regional perfusionabnormalities. J Cardiovasc Magn Reson 2004;6:803–9.
60. Takase B, Nagata M, Kihara T, Kameyawa A, Noya K, Matsui T,et al. Whole-heart dipyridamole stress first-pass myocardial perfu-sion MRI for the detection of coronary artery disease. Jpn Heart J2004;45:475–86.
61. Wolff SD, Schwitter J, Coulden R, Friedrich MG, Bluemke DA,Biederman RW, et al. Myocardial first-pass perfusion magneticresonance imaging: a multicenter dose-ranging study. Circulation2004;110:732–7.
62. Dornier C, Ivancevic MK, Thevenaz P, Vallee JP. Improvementin the quantification of myocardial perfusion using an auto-matic spline-based registration algorithm. J Magn Reson Imaging2003;18:160–8.
63. Gupta SN, Solaiyappan M, Beache GM, Arai AE, Foo TK. Fastmethod for correcting image misregistration due to organ motionin time-series MRI data. Magn Reson Med 2003;49:506–14.
64. Elkington AG, He T, Gatehouse PD, Prasad SK, Firmin DN, PennellDJ. Optimization of the arterial input function for myocardial perfu-sion cardiovascular magnetic resonance. J Magn Reson Imaging2005;21:354–9.
65. Kellman P, Aletras AH, Hsu L, McVeigh ER, Arai AE. T2∗ Mea-surement During First-Pass Contrast-Enhanced Cardiac PerfusionImaging. Magn Reson Med 2006;56:1132–4.
66. Cernicanu A, Axel L. Theory-based signal calibration with single-point T1 measurements for first-pass quantitative perfusion MRIstudies. Acad Radiol 2006;13:686–93.
67. Hsu LY, Kellman P, Arai AE. Correction for T1-Nonlinearity in My-ocardial Signal Intensity Improves First-Pass Perfusion Quantifi-cation. Tenth Annual Scientific Sessions of SCMR. 2007; Rome,Italy.
68. Donahue KM, Burstein D, Manning WJ, Gray ML. Studies of Gd-DTPA relaxivity and proton exchange rates in tissue. Magn ResonMed 1994;32:66–76.
